AI Engineer World's Fair 2025
2025-06-03..2025-06-05
San Francisco, United States
[AI Engineer World's Fair 2024, AI Engineer World's Fair 2025, AI Engineer World's Fair 2026]
as the organisers put it
Join 3,000 engineers and leaders building with AI. June 3 - 5, 2025, San Francisco.
who it's for
ai engineers, ml engineers, platform engineers, infrastructure engineers, developer tools engineers, research engineers, product managers, data engineers
topics
Model Context Protocol (MCP) servers; agent architectures and orchestration; retrieval-augmented generation stacks; RAG evaluation without golden answers; fine-tuning and RL for models; sandboxed microVMs for code execution; voice-first conversational overlays; Rust for agentic tooling; search-based deep learning compilers; knowledge graph integration for RAG; multimodal and generalist models (Qwen); developer workflow automation with agents
| title | author | length |
|---|---|---|
| AI Engineer World's Fair 2025 - Day 1 Keynotes & MCP track ft. Anthropic MCP team | Laurie Voss, Shawn Swyx Wang | 08h 38m |
Explains why MCP exists: a protocol for model agency that Anthropic open‑sources (November launch) so LLMs can call real tools. Covers concrete work shipped at… | ||
| Why the Best AI Agents Are Built Without Frameworks (Primitives over Frameworks) | Ahmad Awais | 27m |
Claims that frameworks slow iteration and primitives win, backed by concrete results: thousands of agents scaled to 100 million monthly runs, eight patterns… | ||
| Rust is the language of the AGI | Michael Yuan | 29m |
Claims that Rust fits AI coding and backs it with a concrete system: the open-source Rust Coder implements an MCP agentic framework that compiles and executes… | ||
| The 4 Patterns of AI Native Development | Patrick Debois | 14m |
Claims four concrete shifts: producer to manager where agents generate code and humans review with diagram-aware, moldable IDE views and guarded permissions… | ||
| Arrakis: How To Build An AI Sandbox From Scratch | Abhishek Bhardwaj | 40m |
Arrakis is an open source microVM sandbox that claims secure, fast code execution with boots in less than 7 seconds and a PR aiming for under one second, uses… | ||
| MCPs are Boring (or: Why we are losing the Sparkle of LLMs) | Manuel Odendahl | 29m |
Argues MCP/tool-calling is stalling LLM potential and shows concrete fixes: pass chat history/memories/files into tool calls, offer editable tool inputs and… | ||
| The RAG Stack We Landed On After 37 Fails | Jonathan Fernandes | 19m |
A concrete RAG stack emerges after 37 fails: prototype with LlamaIndex in Google Colab and BGE small embeddings from Hugging Face, switch to Quadrant as the… | ||
| Letting AI Interface with your App with MCP | Kent C. Dodds | 20m |
Claim: a standard Model Context Protocol lets assistants call services so one assistant can handle arbitrary web tasks. Demo builds live an MCP server and… | ||
| The Future of Qwen: A Generalist Agent Model | Junyang Lin | 25m |
Qwen3 arrives with concrete claims: a flagship 235B total with ~22B activated dense/MoE mix, a deployable 4B model for mobile, and RL fine-tuning that lifts… | ||
| Exposing Agents as MCP servers with mcp-agent | Sarmad Qadri | 18m |
Claim: agents run as MCP servers so any MCP client (Claude, Cursor) can invoke them. Implementation uses mcp-agent built on Model Context Protocol with… | ||
| ChatGPT is poorly designed. So I fixed it | Ben Holmes | 05m |
Claims a practical fix: a working app that makes ChatGPT voice and text at the same time and multi-modal and multi-model. Demonstrates a demoed UI that shows… | ||
| Blender MCP and The Future Of Creative Tools | Siddharth Ahuja | 17m |
Blender MCP glues LLMs like Claude or Cursor to Blender via the MCP protocol and an add on that executes generated scripts so prompts can spawn models and… | ||
| 7 Habits of Highly Effective Generative AI Evaluations | Justin Muller | 26m |
Claims that evaluations are the missing piece to scale with a concrete customer story: a document‑processing workload goes from 22% to 92% accuracy after… | ||
| GPU-less, Trust-less, Limit-less: Reimagining the Confidential AI Cloud | Mike Bursell | 44m |
Claims concrete guarantees: hardware attestation from TEEs (Intel TDX, AMD SEV, Nvidia GPU TES) with on‑chain order reports, SPCTL tooling, and open-source… | ||
| Are MCPs Overhyped? A Rant about MCPs | Henry Mao | 07m |
Claim: Smithery argues the missing piece is an orchestration layer that gives agents context action and learning. Demonstration shows an agent using MCPs to… | ||
| The Demo I Wish I'd Had: OpenAI's Agents SDK... serverless! | Brook Riggio | 18m |
Concrete recipe and working demo: deploys an OpenAI Agents SDK workflow as Python serverless functions on Vercel, orchestrated by Inngest on Vercel to avoid… | ||
| How to Build Trustworthy AI | Allie Howe | 24m |
Claims that matter: concrete defenses and tools. Notes include the Chevy Tahoe $1 chatbot glitch, Slack prompt‑injection data leaks, and Fortnite’s Darth Vader… | ||
| MCP Agent Fine tuning Workshop | Ronan McGovern | 36m |
Agent runs on MCP servers using a Playwright browser, generates and saves MCP browser tool traces (tools list + multi‑turn messages), pushes an unrolled… | ||
| Analyzing 10,000 Sales Calls With AI In 2 Weeks | Charlie Guo | 10m |
Claims that 10,000 sales calls become analyzable in two weeks using LLMs backed by RAG and prompt caching are quantified here with concrete numbers and… | ||
| Agentic Enterprise - What your CEO must know about AI | Hubert Misztela | 28m |
Claims rapid change: organizations may be largely run by AI agents within three years, agents are LLMs glued to real tools (RAG, GUI navigators, coding… | ||
| Text-to-Speech Data Preparation and Fine-tuning Workshop | Ronan McGovern | 34m |
Workshop shows how to train Sesame's CSM-1B on a voice extracted from a YouTube video, using Whisper transcription to make ~41 clips of up to 30 seconds each… | ||
| Break It 'Til You Make It: Building the Self-Improving Stack for AI Agents | Aparna Dhinakaran | 14m |
Claims a deployed agent needs a continuous monitoring and improvement stack that actually connects evals, tracing, observability, experimentation and… | ||
| Breaking the Chain: Agent Continuations for Resumable AI Workflows | Greg Benson | 27m |
Claim: an agent can capture its full call stack—tools, goals, partial responses—into a compact JSON blob combined with the messages array to form a… | ||
| The Coherence Trap: Why LLMs Feel Smart (But Aren’t Thinking) | Travis Frisinger | 21m |
Claims that LLMs aren’t thinking and instead use coherence reconstruction: latent coherence pulls language into aligned patterns, not understanding. Evidence… | ||
| The Knowledge Graph Mullet: Trimming GraphRAG Complexity | William Lyon | 33m |
Claims that a property graph front with RDF triple storage back simplifies GraphRAG by keeping a property-model query surface while using triples for scalable… | ||
| open-rag-eval: RAG Evaluation without "golden" answers | Ofer Mendelevitch | 05m |
Claims automated RAG evaluation without golden answers by using LLM judges and open tooling; framework integrates UMBRELA for reference‑free retrieval scoring… | ||
| Beyond Conversation: Why Documents Transform Natural Language into Code | Filip Kozera | 11m |
Claim: structured, document-based natural language becomes a programming abstraction that forces clarity, versioning, and reusable protocols where chat… | ||
| The Many Ends of Programming | Ray Myers | 27m |
Six concrete outcomes and a few measured surprises decide whether to watch: evidence of extreme completion with Cursor and an Open Hands agent producing a… | ||
| Effective AI Agents Need Data Flywheels, Not The Next Biggest LLM | Sylendran Arunagiri | 17m |
Claims a working system that uses Nemo microservices and Nvidia NIM to run a continuous data flywheel that curates feedback, retrains, evaluates and promotes… | ||
| Supercharging developer workflow with Amazon Q Developer | Vikash Agrawal | 13m |
Amazon Q Developer generates a working 2048 game from scratch using Python, FastAPI and Poetry via the CLI and IDE, creates feature code and a README, and… | ||
| Why Bolt.new Won and Most DevTools AI Pivots Failed | Victoria Melnikova | 06m |
Summary of battle-tested patterns from interviews with 17 developer tools startups that actually tried adding AI and either survived or burned out: a… | ||
| The Voice-First AI Overlay: Designing Conversational Co-Pilots | Gregory Bruss | 17m |
An AI overlay listens passively to live calls, scrapes messy captions and runs a lightweight LM pipeline with caption scraping, smart debouncing, and separate… | ||
| Just do it. (let your tools think for themselves) | Robert Chandler | 07m |
Claim: thin MCP wrappers fail; agentic MCPs succeed by offloading reasoning to richer tools. Demonstration uses the Wordware MCP toolbox wired to Claude, a… | ||
| The Agent Native Company | Rick Blalock | 21m |
Claims a startup built 14 months of product in 8 weeks with six people and a stack of agents, cites agents writing changelogs and creating PRs overnight, and… | ||
| Real AI Agents Need Planning, Not Just Prompting | Yuval Belfer | 08m |
Claim: most agents are just prompted LLMs and fail without planning. Built system: AI21 Mastro combines a planner plus a smart execution engine that generates… | ||
| Building Reliable Support Agents Using the Effect Typescript Library | Michael Fester | 07m |
Claims production-ready support agents built with the Effect TypeScript library using a React front end, an internal effect RPC server, effect HTTP, and… | ||
| RAG Evaluation Is Broken! Here's Why (And How to Fix It) | Yuval Belfer, Niv Granot | 11m |
Claim: current RAG benchmarks reward brittle systems that score high but fail on real tasks. Evidence: a 22‑document FIFA corpus yields only 5% and 11% correct… | ||
| Designing AI To Scale Human Thought | Jun Yu Tan | 12m |
A framework for interfaces that blind spot detection flagging proved bugs and auto-generated unit tests, 43% of PRs caught and almost a thousand new tests in… | ||
| The Robots are coming for your job, and that's okay | Elmer Thomas, Maria Bermudez | 09m |
A small docs team automates repetitive work and measures what actually changes: six single-purpose agents behind a Next.js UI that handles grammar, alt text… | ||
| Luminal - Search-Based Deep Learning Compilers | Joe Fioti | 25m |
Search-first deep learning compiler that represents models as graphs of 12 primitive ops and compiles by enumerating equivalent GPU kernels in an e-graph… | ||
| The Benchmarks Game: Why It's Rigged and How You Can (Really) Win | Darius Emrani | 11m |
Benchmarks control billions and the talk exposes three concrete cheat codes: Cherry-picking comparisons with xAI's Grok‑3 graphs, Buying privileged access via… | ||
| Grounded Reasoning Systems for Cloud Architecture | Iman Makaremi | 26m |
Built an AI copilot that uses LangGraph, Flyte and AWS Bedrock (Claude 3) to orchestrate many specialized agents and a domain knowledge graph as long‑term… | ||
| Agents reported thousands of bugs, how many were real? | Ian Butler, Nick Gregory | 19m |
A new SM-100 benchmark uses 100 triaged, validated bugs from 84 public repos to test agentic maintenance, and a new agent called Bismuth leads the… | ||
| Unlocking Africa's Potential with AI | Thabang Ledwaba | 26m |
Claims that AI can auto-initiate ID applications to eliminate lifelong queues and shows usage data — Kenya ranks third for daily ChatGPT use and forecasts… | ||
| From PM at Stripe to Building an AI startup, a recent founder's journey | Mounir Mouawad | 12m |
Project launches an open source SDK for building production agents with a focus on regulated industries and goes public in June last year. Prior work includes… | ||
| The Current State of Browser Agents | Jerry Wu, Wyatt Marshall | 21m |
Web Bench benchmarks browser agents across 450+ sites with ~5,000 read and write tasks and shows concrete numbers: about 80% success on read tasks, fully… | ||
| How agents broke app-level infrastructure | Evan Boyle | 14m |
LLMs force app latencies into the multi-second range so existing serverless expectations break: p1 latency of a couple of seconds and many serverless providers… | ||
| The End of Awkward AI Transcriptions | Travis Bartley, Myungjong Kim | 16m |
Claims concrete results and usable tooling: 6 top-ten models on the Hugging Face ASR leaderboard, a Parakeet2 family for streaming and a Rea Canary family for… | ||
| Invisible Users, Invisible Interfaces: Accelerating Design Iteration with AI Simulation | Alex Liss | 13m |
Argues that using AI simulation speeds needfinding by turning data into active agents: a live data platform feeds Intelligent Twins that run 72 simulated… | ||
| Will Agent evaluation via MCP Stabilize Agent Networks? | Ari Heljakka | 14m |
Claims that Model Contest Protocol (MCP) lets agents self-improve via evaluators that return numeric scores and text explanations; demonstrates a working Root… | ||
| Buy Now, Maybe Pay Later: Dealing with Prompt-Tax While Staying at the Frontier | Andrew Thomspson | 25m |
Agentic product for real‑estate lawyers ships from GPT‑3.5 through GPT‑4 32K to 4 Turbo 40 and 4.1 and mixes system‑one and system‑two models (01 preview, 04… | ||
| My AI Thinks I'm Eating My Feelings (and Other Nutritional Insights) | Rami Alhamad | 10m |
Alma claims personalized nutrition from LLMs plus real user data, shipping a beta under two months after incorporation and a public release in February; it… | ||
| Stop Ordering AI Takeout A Cookbook for Winning When You Build In House | Jan Siml | 11m |
A small in‑house build claims real revenue: two devs and 10+ sprint weeks create an internal GenAI assistant that delivers several million dollars of ARR. The… | ||
| Dispatch from the Future: building an AI-native Company | Dan Shipper | 18m |
Claims a 10x difference between 90% and 100% AI adoption and shows production evidence: 15 people running four real products (Cora, Monologue, Spiral), 7,000… | ||
| AI Consulting in Practice | NLW | 18m |
A consulting‑grounded study collecting roughly 2,500 use cases via roervey.ai and experience from Super.ai finds concrete signals: KPMG survey shows agents in… | ||
| From Vibe Coding To Vibe Engineering | KITZE | 25m |
Argues that LLMs shift front-end work from typing to directing agents and coins vibe engineering as supervising agents rather than hand-coding. Cites Composer… | ||
| VoiceVision RAG - Integrating Visual Document Intelligence with Voice Response | Suman Debnath | 01h 24m |
Claims that Colpali generates multi-vector embeddings directly from document pages to bypass OCR and that late-interaction patch embeddings (token×patch dot… | ||
| Government Agents: AI Agents Meet Tough Regulations | Mark Myshatyn | 17m |
Los Alamos builds an agent that reads papers, generates hypotheses and writes simulation code, then executes those thermodynamic and hydrodynamic runs on… | ||
| Challenges in High Performance Robotics Systems | Rishabh Garg | 13m |
Designed toy robot runs a 2 ms policy loop over a CAN bus carrying ten 100-bit messages; at 1 Mbps that yields ~1 ms of transmission time, explaining the… | ||
| Building an Agentic Platform | Ben Kus | 19m |
Box moves from single-shot LLM extraction to an agentic-first design that orchestrates OCR, multi-model voting (OpenAI, Anthropic, Gemini, Llama), page-image… | ||
| Five hard earned lessons about Evals | Ankur Goyal | 20m |
Eval competence is measured: an eval that lets a team launch update within 24 hours after a new model proves its value, and a clear pipeline that turns a user… | ||
| Perceptual Evaluations: Evals for Aesthetics | Diego Rodríguez | 16m |
Argues that common metrics like FID break when images carry JPEG artifacts because models and training data inherit compressed internet artifacts, so quality… | ||
| How BlackRock Builds Custom Knowledge Apps at Scale | Vaibhav Page, Infant Vasanth | 19m |
BlackRock describes a Kubernetes-native AI framework with a sandbox and app factory that compresses complex document-extraction app delivery from about 3–8… | ||
| Form factors for your new AI coworkers | Craig Wattrus | 16m |
Flatfile builds an AI stack called AI Transform where agents silently generate personalized demo apps, analyze columns for fixes, run code to transform… | ||
| Fuzzing in the GenAI Era | Leonard Tang | 19m |
Claims a production-ready evaluation loop built from Haizing fuzzing plus agent judges, not static datasets: uses an agent library Verdict and RL (GPO) tuning… | ||
| Multi Agent AI and Network Knowledge Graphs for Change | Ola Mabadeje | 19m |
A production pipeline links ServiceNow tickets to a five‑agent system that consults a layered network knowledge graph implemented on ArangoDB using OpenConfig… | ||
| Wisdom-Driven Knowledge Augmented Generation at Scale | Chin Keong Lam | 19m |
Claims that a Wisdom-Driven Knowledge Graph boosts quantitative Knowledge-Augmented Generation vs vector RAG, implemented as a graph DB + Cypher backend with… | ||
| The Next Unicorns: 7 Top AI startups from the HF0 Residency | Diego Rodríguez, Jonas Bauer, Alex Atallah | 22m |
Seven startups pitch concrete wins: an AI creative suite already generating a million images a day and turning a signup into a paid broadcast in two days; an… | ||
| #define AI Engineer | Greg Brockman, Jensen Huang | 41m |
Career narrative centers on repeated independent study, building a GPU rig for ML experiments, and shifting from web startups to AGI-driven work with a… | ||
| The Future of Evals | Ankur Goyal | 05m |
Braintrust reports the average org runs 13 evals a day and some customers run over 3,000 evals a day; eval work is currently manual but Loop automatically… | ||
| Designing AI-Intensive Applications | swyx | 13m |
Argues AI apps shift from 1:1 to 1:100 to 0:n human input:LLM-call ratios, names LM OS and LM SDLC as candidate standard models, and highlights MCP and agents… | ||
| How to look at your data | Jeff Huber, Jason Liu | 19m |
Argues for fast evals: cheap golden query–document pairs that run in seconds to measure retrieval and compare embedding models on real data. Shows a Weights &… | ||
| On Engineering AI Systems that Endure The Bitter Lesson | Omar Khattab | 19m |
Argues that prompts are a horrible abstraction for engineering because they entangle task definition, formatting, and inference strategy, and that premature… | ||
| Evals Are Not Unit Tests | Ido Pesok | 15m |
Built a Fruit Letter Counter on V0 using the AI SDK and GitHub sync, hits GPT‑4.1 variability: it returns the right answer 10/10 then later fails a real user’s… | ||
| 2025 is the Year of Evals! Just like 2024, and 2023, and … | John Dickerson | 19m |
Claim: enterprise evaluation becomes urgent in 2025 because models move from scoring to acting autonomously, budgets unlock after ChatGPT on Nov 30, 2022, and… | ||
| Vibe Coding with Confidence | Itamar Friedman | 21m |
Soft-launch CLI goes live five minutes before the demo and is used to orchestrate multi-agent workflows that call MCP and A2A for agent-to-agent handshakes; a… | ||
| Full Workshop: Realtime Voice AI | Mark Backman | 01h 10m |
Pipecat is an orchestration-first, open source Python framework that wires transports, STT, LLMs and TTS into real‑time voice agents; the workshop shows a… | ||
| Vision AI in 2025 | Peter Robicheaux | 17m |
Argues that current vision models “aren’t smart” and shows concrete fixes: swapping a Dinov2 backbone into a transformer detector yields +5 mAP vs the… | ||
| Practical tactics to build reliable AI apps | Dmitry Kuchin | 15m |
Shows a concrete workflow: start POC that works ~50% of the time, build reverse-engineer real-world evals before iterating, use o3/LLMs to generate 50+… | ||
| How to Improve your Vibe Coding | Ian Butler | 08m |
A large benchmark of agentic coding tools reports stark failures: three of six agents show 10% true positive rate or less across 900+ bug reports, one agent… | ||
| Vibes won't cut it | Chris Kelly | 16m |
Argues that AI can generate lots of code but 30% generated code is not a solution for production systems because that code still runs in complex, failure-prone… | ||
| Real World Development with GitHub Copilot and VS Code | Harald Kirschner, Christopher Harrison | 01h 20m |
Shows GitHub Copilot inside VS Code creating a working React Vite hydration tracker (Material and Fluent variants), scaffolding, installing packages, starting… | ||
| Building Agents at Cloud Scale | Antje Barth | 19m |
Demo shows a code-assistant agent shipped to CLI in 3 weeks, using an open-source Strand agents SDK that defaults to Amazon Bedrock but also plugs into OpenAI… | ||
| State of Startups and AI 2025 | Sarah Guo | 24m |
Prediction: many companies move from co‑pilot augmentation to agentic automation, with agents up 50% among startup applications and early wins in production… | ||
| Useful General Intelligence | Danielle Perszyk | 20m |
Claims a practical agent stack: a Nova foundation model fine-tuned for UI interaction plus an SDK called Nova Act that translates natural language into clicks… | ||
| The 2025 AI Engineering Report | Barr Yaron | 13m |
Survey of 500 AI practitioners finds broad use of LLMs: 94% apply them to two or more use cases and 82% to three or more, with code generation and content… | ||
| Agents vs Workflows: Why Not Both? | Sam Bhagwat | 16m |
Argues the agent vs workflow debate is overstated and shares concrete patterns: teams build pure workflow graphs for repeatable pipelines, others use… | ||
| Why We Don’t Need More Data Centers | Dr. Jasper Zhang | 14m |
Claims that a GPU marketplace plus a global orchestration layer replaces much of new data center buildout: current US capacity is 55 GW, demand hits ~219 GW by… | ||
| Infrastructure for the Singularity | Jesse Han | 20m |
Morph Labs unveils Infinibranch, a virtualization layer that snapshots, branches and replicates full VMs in milliseconds with Docker‑layer caching semantics so… | ||
| Hacking the Inference Pareto Frontier | Kyle Kranen | 20m |
NVIDIA Dynamo is an open source datacenter inference framework that combines disaggregation, KV routing and inference-time scaling to reshape the… | ||
| Pipecat Cloud: Enterprise Voice Agents Built On Open Source | Kwindla Hultman Kramer | 27m |
Pipecat delivers an open source, vendor‑neutral framework and a Pipecat Cloud layer for deploying real‑time voice agents with telephony and WebRTC support… | ||
| [Full Workshop] Building Conversational AI Agents | Thor Schaeff | 01h 02m |
Hands-on walkthrough of a pipeline that transcribes speech, routes text to an LLM, and streams back synthetic speech with live language switching using 11 Labs… | ||
| From Self-driving to Autonomous Voice Agents | Brooke Hopkins | 20m |
Claim: apply Waymo-style evaluation to voice agents by using large-scale simulation and probabilistic, reference-free metrics. Built a CI/CD loop with… | ||
| Why ChatGPT Keeps Interrupting You | Dr. Tom Shapland | 27m |
Voice agents use a simple VAD plus a 0.5s silence rule today, which causes most interruptions. LiveKit’s transformer-based text-based semantic model feeds the… | ||
| Your realtime AI is ngmi | Sean DuBois, Kwindla Kramer | 17m |
Argues that realtime voice AI must be built from the network layer up, not from the model or app, and demonstrates using WebRTC vs WebSockets why WebRTC… | ||
| Serving Voice AI at $1/hr: Open-source, LoRAs, Latency, Load Balancing | Neil Dwyer | 16m |
Deploys Orpheus (Emotive, Realtime TTS) on GPUs with batch inference and LoRA clones, using Orpheus (Emotive, Realtime TTS) and vLLM with FP8 dynamic… | ||
| How to defend your sites from AI bots | David Mytton | 20m |
Nearly half of web requests are automated and in gaming reach about 60%, with specific examples: GPTBot causes 24% of Diaspora traffic and Read the Docs cuts… | ||
| The Unofficial Guide to Apple’s Private Cloud Compute | Jmo | 21m |
Apple rolled out Private Cloud Compute to millions of devices in October 2024 and claims it delivers stateless computation with cryptographic guarantees by… | ||
| How to Secure Agents using OAuth | Jared Hanson | 19m |
Argues that long‑lived API keys must replace API keys with OAuth and shows concrete fixes: MCP’s first draft ships with no auth, the followup spec separates an… | ||
| How we hacked YC Spring 2025 batch’s AI agents | René Brandel | 18m |
Casco reports they compromise 7 of the 16 agents in YC X25, each within 30 minutes, leaking user system prompts and emails, executing arbitrary code to… | ||
| OpenAI on Securing Code-Executing AI Agents | Fouad Matin | 14m |
OpenAI describes a concrete stack for safe code‑executing agents: an agent runs in a dedicated container that produces a PR, uses Codeex CLI and a local_shell… | ||
| Evaluating AI Search: A Practical Framework for Augmented AI Systems | Danna Emmery, Mara Sher | 21m |
Introduces an open-source dynamic eval agent built with Langraph that generates web-grounded Q&A, refreshes datasets continuously, and logs runs in lengthmiss… | ||
| Scaling Enterprise-Grade RAG: Lessons from Legal Frontier | Calvin Qi, Chang She | 17m |
Harvey builds a RAG stack that combines sparse and dense retrieval, domain-specific reranking, and strict privacy controls, running on LanceDB’s AI-native… | ||
| Building Alice’s Brain: an AI Sales Rep that Learns Like a Human | Sherwood, Satwik | 22m |
An AI SDR called Alice that learns by ingesting sellers’ PDFs, websites, calls and videos is built end to end: vendor parsing with Llama Parse, Firecrawl… | ||
| Layering every technique in RAG, one query at a time | David Karam | 20m |
Starts with in-memory embeddings and moves stepwise through BM25, relevance embeddings, cross-encoder re-rankers, custom embeddings and domain ranking signals… | ||
| Building a Smarter AI Agent with Neural RAG | Will Bryk | 19m |
Live code builds two agent apps that retrieve from the web via Exa: one using keyword RAG (classic inverted-index search) and one using neural RAG via vector… | ||
| [Full Workshop] Building Metrics that actually work | David Karam | 40m |
Workshop shows a concrete scoring system: start with 5–10 simple signals then iterate to hundreds (Google Search used ~300) and calibrate against a… | ||
| Make your LLM app a Domain Expert: How to Build an Expert System | Christopher Lovejoy | 19m |
Claims SOTA clinical reasoning and production deployment for health insurers covering 50 million lives by building an adaptive system rather than swapping… | ||
| Shipping Products When You Don't Know What they Can Do | Ben Stein | 20m |
A platform called Teammates builds an autonomous coworker (Stacy) that lives in Gmail and Slack and surprisingly handles being tagged in a Google Doc comment —… | ||
| Shipping something to someone always wins | Kenneth Auchenberg | 16m |
Core claim: ship something to someone because rapid, real-user feedback beats big-bang plans. Build a skateboard-to-car progression, draft the launch blog/FAQ… | ||
| Why your product needs an AI product manager, and why it should be you | James Lowe | 19m |
Argues that teams need an AI product manager with technical depth to resolve AI uncertainty early, showing the incubator built themefinder (outputs comparable… | ||
| Everything is ugly, so go build something that isn't | Raiza Martin | 25m |
Builds NotebookLM (originally Tailwind) from the ground up, ships an auto‑summary that makes 50‑file workflows usable, and finds 90% of first queries are… | ||
| Building the platform for agent coordination | Tom Moor | 20m |
Linear migrates embeddings from PG vector to Coher and rebuilds search with a hybrid index using Turppuffer, then layers a reranking + query‑rewrite pipeline… | ||
| What Is a Humanoid Foundation Model? An Introduction to GR00T N1 | Annika, Aastha | 18m |
NVIDIA’s GR00T N1 is an open Vision‑Language‑Action foundation model that trains end‑to‑end on human videos, robot trajectories and synthetic sims and runs on… | ||
| Real-time Experiments with an AI Co-Scientist | Stefania Druga | 19m |
Demo combines a low-cost, portable lab (under $300) built with micro:bit + JackDAC, webcams and a mobile REC camera tracking running on-device models, a React… | ||
| Scaling AI Agents Without Breaking Reliability | Preeti Somal | 15m |
Temporal claims to outsource reliability and orchestration for agentic AI by providing language‑idiomatic SDKs, a workflow abstraction that encodes retries and… | ||
| Ship Agents that Ship: A Hands-On Workshop | Kyle Penfound, Jeremy Adams | 01h 21m |
Builds a sandboxed, production-minded AI developer agent using a Dagger workspace module plus real LLMs (Claude/GPT), languages (Python/Go/TS), and… | ||
| The AI Engineer’s Guide to Raising VC | Dani Grant, Chelcie Taylor | 34m |
Shows exact templates and artifacts: real cold emails that got replies, real term sheet language, and an original seed deck at deck.jam.dev; claims that… | ||
| Strategies for LLM Evals (GuideLLM, lm-eval-harness, OpenAI Evals Workshop) | Taylor Jordan Smith | 32m |
Walks through hands-on evals that actually run: deploys an IBM-granite model to a VLM inference container on L4 GPUs, then sweeps request rates with GuideLLM… | ||
| Why you should care about AI interpretability | Mark Bissell | 21m |
Mechanistic interpretability demonstrates programmable control of neurons: the Golden Gate demo locates a bridge feature, flipping it forces the model to… | ||
| Information Retrieval from the Ground Up | Philipp Krenn | 01h 48m |
Search sits under RAG as the retrieval layer and the demo shows concrete tradeoffs: Elasticsearch uses an inverted index with BM25 TF‑IDF and token analysis… | ||
| Introduction to LLM serving with SGLang | Philip Kiely, Yineng Zhang | 44m |
SGLang is an open-source serving framework that claims trillions of tokens per day in production and shows day‑zero support for models like DeepSeek and Qwen… | ||
| Robotics: why now? | Quan Vuong, Jost Tobias Springberg | 18m |
Physical Intelligence demonstrates a pipeline of teleoperated data collection with leader-arm teleoperation and cloud annotation that yields 10,000 hours of… | ||
| Waymo's EMMA: Teaching Cars to Think | Jyh Jing Hwang | 17m |
Waymo presents EMMA, a Gemini-based end-to-end multimodal planner that ingests eight-camera video plus routing text and directly predicts future waypoints; it… | ||
| A2A & MCP Workshop: Automating Business Processes with LLMs | Damien Murphy | 01h 23m |
Shows a working webhook-to-agent pipeline that spawns A2A agents on incoming webhooks, uses Google's Agent-to-Agent protocol plus Zapier MCP and a Gemini API… | ||
| Ship Production Software in Minutes, Not Months | Eno Reyes | 16m |
Claims a platform (Factory) that runs thousands of autonomous agents to orchestrate the whole SDLC, ingesting tasks, searching a 10TB codebase, referencing org… | ||
| Beyond the Prototype: Using AI to Write High-Quality Code | Josh Albrecht | 18m |
Imbue describes Sculptor as an experimental coding agent environment that actually enforces quality: it autoplan prompts to force planning before code, detects… | ||
| Software Development Agents: What Works and What Doesn't | Robert Brennan | 17m |
Demonstrates concrete workflows using OpenHands (fka OpenDevin): agents run inside a Docker container, use a diff-based editor and accessibility… | ||
| Devin 2.0 and the Future of SWE | Scott Wu | 16m |
Claims that code agent capability doubles about every 70 days, enabling 16–64x yearly gains, and shows concrete progress from line completion to full project… | ||
| Your Coding Agent Just Got Cloned And Your Brain Isn't Ready | Rustin Banks | 14m |
Google Labs releases Jules, a cloud-first asynchronous coding agent that clones a repo into its own VM, runs tests, and submits PRs; it uses Gemini 2.5 Pro and… | ||
| Latent Space Paper Club: AIEWF Special Edition (Test of Time, DeepSeek R1/V3) | VIbhu Sapra | 54m |
DeepSeek R1 (May 28 update) claims large reasoning gains by extending test‑time compute and pure RL: AIM score rises 70→87.5 and average reasoning traces… | ||
| Human seeded Evals | Samuel Colvin | 12m |
Introduces using Pydantic models as executable contracts so the LLM returns structured output, then feeds validation-as-feedback back into the model to correct… | ||
| Building AI Products That Actually Work | Ben Hylak, Sid Bendre | 19m |
Claim: many popular eval techniques don't predict real-world performance, and reliable AI products require live signals not just offline scores. Built Raindrop… | ||
| Rise of the AI Architect | Clay Bavor, Alessio Fanelli | 19m |
Sierra claims enterprises will ship branded AI agents as the next customer interface and reports hundreds of customers and serving hundreds of millions of… | ||
| AI That Pays: Lessons from Revenue Cycle | Nathan Wan | 18m |
Healthcare revenue cycle drives severe losses — 40% of hospitals operate at a negative margin — and much waste comes from avoidable denials and manual prior… | ||
| Structuring a modern AI team | Denys Linkov | 18m |
Argues that most AI problems need people, not just models, and proposes hiring generalists who can fine-tune encoders, run Hugging Face pipelines, and ship… | ||
| The Rise of Open Models in the Enterprise | Amir Haghighat | 17m |
Calls out why enterprises move from API-driven OpenAI/Anthropic builds to self-hosted open models: quality wins on domain tasks (medical extraction, jargon… | ||
| Mentoring the Machine | Eric Hou | 19m |
Claims that AI agents can turn weeks of work into hours are backed here by concrete results: an Augment Extension agent finds and reverts a bad commit across… | ||
| Building Applications with AI Agents | Michael Albada | 16m |
Generative models can reach ~70% out-of-the-box on complex agentic tasks but climbing past that long tail requires architecture and evaluation: register few… | ||
| AX is the only Experience that Matters | Ivan Burazin | 15m |
Claims that 25% of recent YC startups have AI writing 95% of their code and that 37% build agents drive a focus on tooling. Daytona presents an agent-native… | ||
| How to build Enterprise Aware Agents | Chau Tran | 20m |
Argues for a practical Workflow Search paradigm that turns agent traces into reusable, evaluable workflows and backfills enterprise context into LLM loops… | ||
| Monetizing AI | Alvaro Morales | 18m |
Explains why traditional SaaS pricing fails for AI and shows a working tool that simulates pricing on real product usage: Orb Simulations ingests real-time… | ||
| Does AI Actually Boost Developer Productivity? (100k Devs Study) | Yegor Denisov-Blanch | 18m |
Stanford analyzes nearly 100,000 developers across hundreds of companies and finds AI yields an average productivity boost of ~20% but with wide variance and… | ||
| How agents will unlock the $500B promise of AI | Donald Hruska | 16m |
Retool launches Retool agents last week as a production-focused agent platform that blends LLM loops with tool calls, guardrails, SSO, RBAC, audit logs and… | ||
| How Intuit uses LLMs to explain taxes to millions of taxpayers | Jaspreet Singh | 19m |
Intuit uses Anthropic and OpenAI models plus a proprietary Geno platform to generate TurboTax explanations for 44 million returns, combining fine-tuned Claude… | ||
| 3 ingredients for building reliable enterprise agents | Harrison Chase | 21m |
Three concrete levers raise expected value for enterprise agents: pick high‑value tasks or shift to longer, ambient workflows so agents do substantial work… | ||
| From Hype to Habit: How We’re Building an AI-First SaaS Company—While Still Shipping the Roadmap | Deepsha Menghani | 19m |
Sprout Social describes rearchitecting strategy, systems, teams and incentives to put AI at product center and moves from deterministic roadmaps to ritualized… | ||
| Machines of Buying and Selling Grace | Adam Behrens | 20m |
Claims that AI will replace browsing with merchant and consumer agents that negotiate intent, using programmatic access or agent-friendly sites; demonstrates a… | ||
| How to Build Planning Agents without losing control | Yogendra Miraje | 16m |
Describes a working plan-and-execute pipeline built on an LLM compiler pattern with blueprint generator, planner, executor and joiner nodes implemented on… | ||
| Building Agents (the hard parts!) | Rita Kozlov | 21m |
Claims that agents move work from augmentation to full automation with measured wins — 20% revenue uplift, 90% faster support and 50–75% time savings — and… | ||
| POC to PROD: Hard Lessons from 200+ Enterprise GenAI Deployments | Randall Hunt | 19m |
Practices from 200+ deployments show concrete tradeoffs: prompt engineering often outperforms fine-tuning now; embeddings alone do not make a faceted query… | ||
| Build Dynamic Products, and Stop the AI Sideshow | Eliza Cabrera, Jeremy Silva | 18m |
Argues for a crawl walk run path with concrete milestones: semantic search on a shared inbox (crawl), draft suggestions and contextual co‑pilot surfacing… | ||
| The Billable Hour is Dead; Long Live the Billable Hour | Kevin Madura, Mo Bhasin | 17m |
An internal GenAI platform with 20 engineers reaches 50+ deployments and cuts tasks from days to minutes by using structured outputs, retrieval-augmented… | ||
| From Copilot to Colleague: Trustworthy Agents for High-Stakes | Joel Hron | 20m |
Claims a move from helpful assistants to agentic systems that actually produce judgments and end to end work using legacy tools, backed by 4,500 domain experts… | ||
| How to Hire AI Engineers when EVERYONE is cheating with AI | Beth Glenfield | 07m |
Claims that AI recruitment is broken with concrete numbers: Clu raises $5.3M and heads toward $1M ARR, hiring interviews show 93% LeetCode success rates and… | ||
| Stateful environments for vertical agents | Josh Purtell | 07m |
Repository implements a stateful environment layer that gives games and apps a same interface for Pokemon Red Minecraft Swe-Bench, isolates logic into a… | ||
| Books reimagined: AI to create new experiences for things you know | Lukasz Gandecki | 10m |
A prototype reads full books, tags scenes with LLM scene analysis and character detection, generates music and avatars from mood prompts, and outputs… | ||
| AI powered entomology: Lessons from millions of AI code reviews | Tomas Reimers | 10m |
Analysis of millions of automated code reviews from Graphite’s Diamond shows LLMs can reliably flag certain bug classes but also flood PRs with unwelcome… | ||
| Critical AI Inference your CIO can Trust | Sahil Yadav, Hariharan Ganesan | 19m |
Claims a practical ops layer called XTOPS that adds explainability, traceability and dynamic guardrails to MLOps and cuts incident turnaround from months to… | ||
| How to run Evals at Scale: Thinking beyond Accuracy or Similarity | Muktesh Mishra | 09m |
Argues that evaluations must measure actionable things not just accuracy, showing concrete practices: start with synthetic data then expand to multiple labeled… | ||
| Continuous Profiling for GPUs | Matthias Loibl | 12m |
Continuous profiling extends an existing platform to deliver always-on visibility into GPU behavior by sampling with Linux EVPF and querying NVIDIA NVML; it… | ||
| Top Ten Challenges to Reach AGI | Stephen Chin, Andreas Kollegger | 04m |
A rapid opener lists ten sci-fi memes and maps each to a concrete engineering challenge — short‑term memory to prompt engineering, HAL to transparency and… | ||
| Practical GraphRAG: Making LLMs smarter with Knowledge Graphs | Michael, Jesus, Stephen | 20m |
Shows a concrete GraphRAG pipeline: ingest PDFs, transcripts and web pages into a lexical graph, extract entities with LLM prompts, run graph algorithms for… | ||
| Knowledge Graphs in Litigation Agents | Tom Smoker | 19m |
Company pipeline scrapes the web, extracts and schemas data into knowledge graphs, and runs multi-agent workflows with human-in-the-loop checks to find… | ||
| When Vectors Break Down: Graph-Based RAG for Dense Enterprise Knowledge | Sam Julien | 16m |
Claims 86.31% accuracy on RobustQA with sub‑second responses and outperforms seven vector systems; uses a small specialized LLM to map documents into… | ||
| Stop Using RAG as Memory | Daniel Chalef | 07m |
Argues that vectorstore RAG fails as memory and shows a working alternative: temporally-aware knowledge graphs built with the open-source Graphiti/Zep stack… | ||
| HybridRAG: A Fusion of Graph and Vector Retrieval | Mitesh Patel | 20m |
Introduces a HybridRAG pipeline that fuses knowledge-graph triplets with vector embeddings, uses LLM prompt-engineering and fine-tuning, and evaluates… | ||
| tldraw.computer | Steve Ruiz | 19m |
A visual programming canvas where arrows connect LLM-powered blocks, executing multimodal pipelines that produce text, images and speech in sequence; the… | ||
| Excalidraw: AI and Human Whiteboarding Partnership | Christopher Chedeau | 17m |
Excalidraw integrates AI by turn prompt into diagram using LLMs to emit a structured diagram DSL (like Mermaid) and an implemented Mermaid JS → Excal backend… | ||
| The Bitter Layout or: How I Learned to Love the Model Picker | Maximillian Piras | 14m |
Argues that the ubiquitous input field + turn‑by‑turn flow + dropdown of models—the model picker—exists because models are not yet commoditized, citing… | ||
| CIAM for AI: Authn/Authz for Agents | Michael Grinich | 20m |
Claims that agents need first-class, hybrid machine+person identities and that agent traffic will flip from ~5% to ~95%. Lists four practical patterns: persona… | ||
| Good design hasn’t changed with AI | John Pham | 20m |
Covers concrete front-end techniques and measurable UX tradeoffs: server-side rendering, caching, bundle-size targets tied to LCP/INP/CLS, and aiming for first… | ||
| Building Effective Voice Agents | Toki Sherbakov, Anoop Kotha | 17m |
Real-time speech-to-speech API that combines transcription, LLM reasoning, and TTS reduces end-to-end latency and preserves conversational context compared… | ||
| Robots as professional Chefs | Nikhil Abraham | 19m |
A bimanual mobile robot learns new recipes from a single expert demonstration and runs in real commercial kitchens, achieving 95% autonomous operation with… | ||
| [Full Workshop] Reinforcement Learning, Kernels, Reasoning, Quantization & Agents | Daniel Han | 02h 42m |
Claims that RL can replace stages of SFT, plus a concrete algorithmic tweak: GRPO removes the value model and uses rollout statistics as the baseline, letting… | ||
| Google Photos Magic Editor: GenAI Under the Hood of a Billion-User App | Kelvin Ma | 20m |
Google Photos integrates traditional CV and large generative models to power features like Magic Eraser and post‑capture portrait segmentation, serving 1.5… | ||
| Dream Machine: Scaling to 1m users in 4 days | Keegan McCallum | 19m |
Luma AI hits a million users in four days by turning a brittle Triton-based stack into a vanilla PyTorch serving stack, manually scaling from 500 to 5,000… | ||
| ComfyUI Full Workshop | ComfyAnonymous | 51m |
Node-based ComfyUI is a locally runnable canvas for image, video, audio, 3D and text generation that claims 3–4 million active users and sits in the top 150… | ||
| Design like Karpathy is watching | Zeke Sikelianos | 19m |
MenuGen is a photo‑from‑menu web app built with generative models that works locally but trips over deployment: docs and Replicate API were outdated, an… | ||
| On Curiosity | Sharif Shameem | 19m |
Argues curiosity is the engine that pulls future ideas into the present and demonstrates it with concrete demos: a 2020 GPT‑3 prototype constrained by a… | ||
| Real world MCPs in GitHub Copilot Agent Mode | Jon Peck | 14m |
Agent mode builds a working app in about eight minutes by reading a detailed README, running commands in VS Code, and asking permission before touching… | ||
| The rise of the agentic economy on the shoulders of MCP | Jan Curn | 18m |
Claims that MCP tool discovery lets agents dynamically find and call services across Apify’s marketplace of ~5,000 “actors,” using one API token to access… | ||
| Full Spec MCP: Hidden Capabilities of the MCP spec | Harald Kirschner | 15m |
VS Code implements the full MCP spec (v1.10 in insiders) and shows concrete features: dynamic tool discovery that exposes battle or debugging tools only when… | ||
| Shipping an Enterprise Voice AI Agent in 100 Days | Peter Bar | 17m |
Finn Voice goes from blank page to a deployed enterprise phone agent in about 100 days, using real‑time APIs and a stack of STT–LLM–TTS + RAG + telephony while… | ||
| The State of Generative Media | Gorkem Yurtseven | 17m |
Company fal runs inference on DALL·E 2, Midjourney, Stable Diffusion/SDXL and newer models like Sora and DeepMind V3 and reports video model usage rising from… | ||
| Teaching Gemini to Speak YouTube: Adapting LLMs for Video Recommendations to 2B+DAU | Devansh Tandon | 23m |
YouTube rebuilds recommendations by adapting Gemini into a domain-specific checkpoint called LRM that learns a new tokenization for videos — SemanticID — via… | ||
| Transforming search and discovery using LLMs | Tejaswi, Vinesh | 21m |
Instacart replaces brittle engagement-trained components with LLM-backed modules for query classification and rewrites, precomputing head/torso outputs and… | ||
| Netflix's Big Bet: One model to rule recommendations | Yesu Feng | 22m |
Claims one shared autoregressive transformer can replace many specialized recommenders and yields measurable gains: scale-up from models trained on ~10 million… | ||
| 360Brew: LLM-based Personalized Ranking and Recommendation | Hamed, Maziar | 22m |
Claims a 150B-parameter foundation model called Blue XL is trained with promptification and instruction fine-tuning and then distilled stepwise to usable sizes… | ||
| What We Learned from Using LLMs in Pinterest | Mukuntha Narayanan, Han Wang | 18m |
Pinterest integrates LLMs into search relevance, fine-tuning a cross-encoder teacher on human labels and using it to label daily logs for semi-supervised… | ||
| Measuring AGI: Interactive Reasoning Benchmarks for ARC-AGI-3 | Greg Kamradt | 18m |
ARC Prize Foundation previews ARC‑AGI‑3: a dynamic, game-like benchmark with ~40 public training games and a hidden private evaluation set (no internet… | ||
| RL for Autonomous Coding | Aakanksha Chowdhery | 19m |
Argues that scaling LLMs with post-training open-weight LLMs plus reinforcement learning measurably raises autonomous coding success, citing PaLM-scale models… | ||
| Recsys Keynote: Improving Recommendation Systems & Search in the Age of LLMs | Eugene Yan | 21m |
Recommendation builds multimodal semantic IDs (Quaisho) that lift cold‑start coverage by 3.6% and raise cold‑start velocity, uses LLMs to synthesize labels… | ||
| Benchmarks Are Memes: How What We Measure Shapes AI—and Us | Alex Duffy | 16m |
Benchmarks act like cultural memes shaping model behavior: Every Inc. builds an AI Diplomacy benchmark and runs 18 models against each other, finding Gemini 25… | ||
| Small AI Teams with Huge Impact | Vik Paruchuri | 18m |
Datalab scales revenue 5x to seven‑figure ARR while keeping headcount at four and 40K GitHub stars for repos surya and marker. A 500M‑parameter OCR model ships… | ||
| Rethinking Team Building: how a 30-person Startup serves 50 Million Users | Grant Lee | 18m |
A 30-person startup scales to 50 million users by hiring generalists who code, design, research and mentor rather than specialists, centering on a player coach… | ||
| Building a 10 person unicorn | Max Brodeur-Urbas | 12m |
Gumloop claims it scales automation for Instacart, Webflow and Shopify while staying at less than 10 people, hitting millions in ARR after YC and a Series A… | ||
| Using OSS models to build AI apps with millions of users | Hassan El Mghari | 19m |
Builds and launches dozens of tiny open‑source AI apps, several hitting millions: roomGPT.io 2.9M, restorePhotos.io 1.1M, Blinkshot.io 1M, LlamaCoder.io 1.4M… | ||
| Bolt.new: How we scaled $0-20m ARR in 60 days, with 15 people | Eric Simons | 18m |
Claimed $0–$20M ARR in 60 days with under 15 people, scaling from a stripped-down MVP race car product while the core team handled support directly. Early… | ||
| Prompt Engineering and AI Red Teaming | Sander Schulhoff | 02h 01m |
Creator of Learn Prompting and HackAPrompt describes concrete results: an open dataset of 600,000 prompts, a 76‑page Prompt Report surveying 1,500+ papers and… | ||
| Survive the AI Knife Fight: Building Products That Win | Brian Balfour | 14m |
Reforge encodes product judgment into AI agents like Reforge Insights (research) and Compass (project manager) and claims differentiation comes from combining… | ||
| Automating Escrow with USDC and AI | Corey Cooper | 58m |
Claims that USDC plus AI can automate escrow by parsing agreements and triggering instant, programmable USDC payouts; demonstrator uses USDC smart contract on… | ||
| How LLMs work for Web Devs: GPT in 600 lines of Vanilla JS | Ishan Anand | 01h 42m |
A full GPT-2 small inference is implemented in 600 lines of Vanilla JS that loads about 1.5 GB of model weights into the browser IndexedDB and runs locally… | ||
| [Workshop] AI Pipelines and Agents in Pure TypeScript with Mastra.ai | Nick Nisi, Zack Proser | 01h 51m |
Mastra.ai shows a pure TypeScript framework that builds typed workflows and agents, runs a local MCP server, and connects OpenAI + imageflip to turn a text… | ||
| AI Engineering with the Google Gemini 2.5 Model Family | Philipp Schmid | 01h 45m |
Hands-on workshop shows live use of Gemini 2.5 Flash (free API tier) and Gemini 2.5 Pro, uploading files to AI Studio, counting tokens (input/output/thought)… | ||
| The New Code | Sean Grove | 22m |
Claim: specifications, not prompts or code, become the fundamental unit of programming. Evidence: OpenAI publishes a living Model Spec as versioned markdown… | ||
| Production software keeps breaking and it will only get worse | Anish Agarwal | 18m |
Claims an autonomous troubleshooting product, Traversal.ai, that combines causal machine learning with LLM reasoning and agent swarms to search petabytes of… | ||
| Thinking Deeper in Gemini | Jack Rae | 18m |
Identifies a test‑time compute bottleneck and adds an explicit thinking stage trained with reinforcement learning so the model can emit iterative tokens and… | ||
| A year of Gemini progress + what comes next | Logan Kilpatrick | 12m |
A new Gemini 2.5 Pro model ships with measurable benchmark gains (improved scores on ADER and HLE) and is positioned as a final 2.5 update, while server… | ||
| 2025 in LLMs so far, illustrated by Pelicans on Bicycles | Simon Willison | 19m |
A rapid six‑month survey counts about 30 significant model releases and introduces a code‑based test: the Pelican Riding a Bicycle benchmark — prompt models to… | ||
| Trends Across the AI Frontier | George Cameron | 18m |
Artificial Analysis benchmarks 150+ models with a seven-evaluation intelligence index and finds clear trade-offs: reasoning models use an order of magnitude… | ||
| Training Agentic Reasoners | Will Brown | 19m |
Argues that RL at scale reliably improves multi-turn tool-using agents and shows a concrete pipeline: SFT warm-up, then RL with GRPO vs PPO tradeoffs for… | ||
| New York Times' Connections: A Case Study on NLP in Word Games | Shafik Quoraishee | 19m |
Builds a pipeline that converts the 16-word board into multi-dimensional semantic graphs using WordNet, ConceptNet and embeddings, scores pairwise relations… | ||
| Claude Code & the evolution of agentic coding | Boris Cherny | 18m |
Claims that the model is improving exponentially and products lag, and Claude Code focuses on a terminal-first integration that runs in any shell, VS Code… | ||
| 12-Factor Agents: Patterns of reliable LLM applications | Dex Horthy | 17m |
Argues for 12 engineering factors (repo: humanlayer/12-factor-agents) that make LLM apps reliable: convert language to JSON and treat tools as structured… | ||
| MCP Is Not Good Yet | David Cramer | 17m |
Claims Sentry built a remote MCP server in a few days, wires it into VS Code Insiders and Cursor via a Cloudflare shim for OAuth 2.1, and exposes curated ops… | ||
| Your Personal Open-Source Humanoid Robot for $8,999 | JX Mo | 19m |
An open-source humanoid platform sells pre-orders at $8,999 with full CAD, PCB, firmware and ML released under MIT, aiming for delivery by October; hardware… | ||
| The Build-Operate Divide: Bridging Product Vision and AI Operational Reality | Chris Hernandez, Jeremy Silva | 13m |
Claim: product progress stalls at a "quality chasm" unless iteration and ops scale; the solution is a unified pipeline from prototyping and prompt tests to… | ||
| The New Lean Startup | Sid Bendre | 13m |
Oleve claims a profitable multi-product portfolio at about $6M ARR with a tiny team of four, two consumer apps totaling over 250M social views, one reaching… | ||
| Optimizing inference for voice models in production | Philip Kiely | 15m |
Demo uses Orpheus TTS (Llama 3.2 3B backbone) served with TensorRT-LLM + FP8 quantization and torch.compile on GPU for audio decoding, achieving ~150 ms TTFB… | ||
| Conquering Agent Chaos | Rick Blalock | 15m |
Platform for deploying stateful agents that runs Bun, Python (uvicorn), and Node.js, wires to Vercel AI SDK and Grok, and auto-deploys from GitHub using a YAML… | ||
| [Evals Workshop] Mastering AI Evaluation: From Playground to Production | Carlos Esteban, Doug Guthrie | 01h 25m |
Hands-on workshop shows how to run evals from playground to production using Braintrust: push prompts and datasets from a repo with pnpm/braintrust push, run… | ||
| Intro to GraphRAG | Zach Blumenfeld | 01h 19m |
Hands-on introduction that builds a talent graph in Neo4j from a CSV and two sample résumés, creates uniqueness constraints, loads ~154 nodes, stores 1536-dim… | ||
| Securing Agents with Open Standards | Bobby Tiernay, Kam Sween | 19m |
Shows a concrete architecture that replaces shared API keys with short-lived per-user tokens minted via token exchange, an open-source Node.js/TypeScript agent… | ||
| The emerging skillset of wielding coding agents | Beyang Liu | 35m |
Claims that agents require new workflows not old prompts and that Amp can autonomously edit real production codebases and run tests is demonstrated with a live… | ||
| Turning Fails into Features: Zapier’s Hard-Won Eval Lessons | Rafal Willinski, Vitor Balocco | 16m |
Zapier Agents turns failures into a data flywheel by instrumenting traces (tool calls, errors, pre/postprocessing) and offering a one‑click path to convert a… | ||
| Building voice agents with OpenAI | Dominik Kundel | 01h 26m |
Claims that voice agents are an API to the real world and that OpenAI now ships an OpenAI Agents SDK for TypeScript with native voice support and guardrails… | ||
| Containing Agent Chaos | Solomon Hykes | 24m |
Argues that isolating every agent step inside containers turns flaky LLM workflows into reproducible, debuggable pipelines, and demos a prototype that wires… | ||
| Evals 101 | Doug Guthrie | 49m |
Walks through building end-to-end evals on Braintrust: define a task (prompt or agent), push datasets and scores (code or LLM-as-judge), run offline… | ||
| Why should anyone care about Evals? | Manu Goyal | 06m |
Evals are positioned as more than unit tests: Braintrust builds a dev platform that runs offline experiments to catch faults before deployment, lets teams… | ||
| To the moon! Navigating deep context in legacy code with Augment Agent | Forrest Brazeal, Matt Ball | 15m |
Augment Agent indexes a legacy Apollo guidance computer repo, uses its context engine to map assembly to meaning, and explains the 1201/1202 alarms as the AGC… | ||
| Serving Voice AI at Scale | Arjun Desai, Rohit Talluri | 17m |
Claims that state space models replace transformers for real‑time voice by keeping generation O(1) while matching transformer quality, with architectures tuned… | ||
| Ship it! Building Production Ready Agents | Mike Chambers | 20m |
A simple D20 agent runs locally on Llama 3.1 8B then scales to a fully managed deployment using Amazon Bedrock agents with a dice-rolling Lambda tool wired… | ||
| Introducing Strands Agents, an Open Source AI Agents SDK | Suman Debnath | 14m |
Open-source Strands Agents claims to cut agent development from months to days by using a model plus tools only; the demo reads a disk file, generates a… | ||
| Data is Your Differentiator: Building Secure and Tailored AI Systems | Mani Khanuja | 20m |
Argues that proprietary data is the competitive layer and shows how Amazon Bedrock components turn that data into secure, production-grade LLM apps: Amazon… | ||
| How to build world-class AI products | Sarah Sachs, Carlos Esteban | 01h 44m |
Notion AI describes running production AI for 100 million users, switching new models into production in under a day, and spending roughly 10% on prompting and… | ||
| From Mixture of Experts to Mixture of Agents with Super Fast Inference | Daniel Kim, Daria Soboleva | 53m |
Hands-on workshop shows how to assemble a Mixture of Agents (MoA) system using Qwen3-32B and Llama 3.3-70B running on Cerebras hardware, claiming MoA can beat… | ||
| Forget RAG Pipelines—Build Production Ready Agents in 15 Mins | Nina Lopatina, Rajiv Shah | 01h 16m |
Demo builds a working Agentic RAG prototype in ~15 minutes using Contextual AI's managed RAG stack, with notebook, API key and a GUI at app.contextual.ai; it… | ||
| Milliseconds to Magic: Real‑Time Workflows using the Gemini Live API and Pipecat | Kwindla Kramer | 22m |
Demonstration shows Gemini Live API GA powered by Gemini 2.5 Flash combined with the Pipecat framework to run real‑time multimodal voice agents handling… | ||
| Realtime Conversational Video with Pipecat and Tavus | Chad Bailey, Brian Johnson | 19m |
Tavus demonstrates a realtime conversational video pipeline that produces synchronized avatar audio+video with ~600 ms response for baseline interactions… | ||
| Vector Search Benchmark[eting] | Philipp Krenn | 14m |
Marketing charts often hide the setup: vendors cherry‑pick a scenario to claim 40% faster overall by promoting one tuned case, and a single edge optimization… | ||
| Taming Rogue AI Agents with Observability-Driven Evaluation | Jim Bennett | 16m |
Claims a repeatable observability pipeline that logs per-step traces, scores them with a stronger LLM, and surfaces actionable fixes; demo chatbot required… | ||
| Building agent fleet architectures your CISO doesn't hate | Lou Bichard | 14m |
Gitpod moves from GCP Kubernetes to a bring your own cloud model that runs a tiny runner as a single ECS task (single‑digit dollars per month) on AWS EC2… | ||
| Don’t get one-shotted: Use AI to test, review, merge, and deploy code | Tomas Reimers | 06m |
AI-generated code now makes up about 46% of GitHub code, and studies warn up to half of AI outputs contain bugs or vulns. Graphite builds Diamond, an AI-native… | ||
| Effective agent design patterns in production | Laurie Voss | 16m |
LlamaIndex is a Python/TypeScript framework with Llama Parse for cleaning PDFs/Word/PowerPoints, a paid Llama Cloud retrieval endpoint, a connector registry… | ||
| Foundry Local: Cutting-Edge AI experiences on device with ONNX Runtime/Olive | Emma Ning | 23m |
Foundry Local packages ONNX Runtime/Olive into an end-to-end on-device stack that Windows and macOS apps can use to run open models locally, claiming 10… | ||
| Unlocking AI Powered DevOps Within Your Organization | Jon Peck | 22m |
Claims measurable gains: teams see up to 1.5x feature throughput and typical improvements around 30%, plus higher test pass rates. Practical tools and files… | ||
| Vibe Coding at Scale: Customizing AI Assistants for Enterprise Environments | Harald Kirshner | 15m |
Shows concrete tooling and workflows: VS Code Insiders with agent mode, a local speech model, and GitHub Copilot driving workspace scaffolds; demo uses Cloudin… | ||
| The Agent Awakens: Collaborative Development with Copilot | Christopher Harrison | 01h 04m |
Explains how GitHub Copilot shifts from line completions to a GitHub-hosted coding agent that runs in GitHub Actions, creates its own branch and creates draft… | ||
| AI Red Teaming Agent: Azure AI Foundry | Nagkumar Arkalgud, Keiji Kanazawa | 20m |
Azure’s Red Teaming Agent in the Azure AI Evaluation SDK runs automated scans against applications or models (RAG on Postgres, local OLLAMA, or Azure OpenAI)… | ||
| Collaborating with Agents in your Software Dev Workflow | Jon Peck, Christopher Harrison | 01h 04m |
Hands-on lab demonstrates GitHub Copilot acting as an agent inside the IDE and on github.com using Model Context Protocol (MCP) to call external tools; agents… | ||
| Agentic Excellence: Mastering AI Agent Evals w/ Azure AI Evaluation SDK | Cedric Vidal | 20m |
Built an agent with the AI toolkit for VS Code that uses Playwright to scrape event pages, extract name/date/location and attendee count (demo shows 269) by… | ||
| Building Code First AI Agents with Azure AI Agent Service | Cedric Vidal | 01h 54m |
Hands-on demo shows a stateful Azure agent built with the Azure AI Foundry Python SDK that generates SQL, queries SQLite sales data, and merges results with… | ||
| How fast are LLM inference engines anyway? | Charles Frye | 16m |
Benchmarks cover hundreds of runs across open models, engines (VLM, SGLang, Tensor TLM), and H100 GPUs with raw data and code posted. Key reproducible claims… | ||
| RAG in 2025: State of the Art and the Road Forward | Tengyu Ma | 19m |
Argues that RAG beats fine-tuning and naive long-context for enterprise use, cites an average retrieval accuracy of ~80% across ~100 datasets with wide… | ||
| The State of AI Powered Search and Retrieval | Frank Liu | 13m |
Overview of practical results and engineering choices for AI search: RAG plus reranking is standard, embeddings drive relevance and embedding quality is… | ||
| Architecting Agent Memory: Principles, Patterns, and Best Practices | Richmond Alake | 18m |
Architecture advice and concrete components for agent memory: a modular lifecycle of generation, storage, retrieval, integration, updating, forgetting… | ||
| Building Multimodal AI Agents From Scratch | Apoorva Joshi | 37m |
Builds a multimodal agent that answers questions over mixed text+images and explains charts, implemented in Python using MongoDB as a vector database and… | ||
| Why Your Agent’s Brain Needs a Playbook: Practical Wins from Using Ontologies | Jesús Barrasa | 14m |
Argues that serializing an ontology into a Neo4j-backed knowledge graph turns GraphRAG from brittle into actionable: built a pipeline that ingests structured… | ||
| Memory Masterclass: Make Your AI Agents Remember What They Do! | Mark Bain | 51m |
Workshop demos show concrete tooling and tradeoffs: four live demos (some fallback to video because of Wi‑Fi) that ingest real data and write stateful graphs… | ||
| Graph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics | Alison, Andreas | 01h 41m |
Shows a reproducible GraphRAG workflow that loads embeddings into Neo4j, builds a vector index on node property embeddings, runs KN&N k=25 to link similar… | ||
| GraphRAG methods to create optimized LLM context windows for Retrieval | Jonathan Larson | 15m |
Demonstrates that structured LLM memory with graphs actually enables repository-level reasoning: graph for code translates a 200-line Python game into working… | ||
| Agentic GraphRAG: Simplifying Retrieval Across Structured & Unstructured Data | Zach Blumenfeld | 15m |
Builds a pipeline that extracts entities from PDFs, loads embeddings and structured ETL into Neo4j graph, uses LangChain decomposition and Pydantic to produce… | ||
| Revenue Engineering: How to Price (and Reprice) Your AI Product | Kshitij Grover | 16m |
Orb builds usage-based billing infrastructure that meters AI workloads, simulates revenue, and supports rapid repricing; example customers moved from… | ||
| "Data readiness" is a Myth: Reliable AI with an Agentic Semantic Layer | Anushrut Gupta | 17m |
Claims a self-improving, AI‑maintained semantic layer that converts messy enterprise data into reproducible queries by generating promql plans (a deterministic… | ||
| Building Agentic Applications w/ Heroku Managed Inference and Agents | Julián Duque, Anush Dsouza | 53m |
Heroku provisions LLMs (Claude 3.5/3.7/4, Cohere embeddings, Stable Image Ultra) as a platform add‑on and runs an agentic control loop that calls tools on… | ||
| Events are the Wrong Abstraction for Your AI Agents | Mason Egger | 15m |
Argues that using Events as the core abstraction fragments business logic and increases latency, then proposes durable execution as the replacement: automatic… | ||
| Prompt Engineering is Dead | Nir Gazit | 14m |
Claims a replaceable workflow: build an automated agent that crawls prompting guides, uses an LM-as-judge evaluator, and iterates prompts against a 20-question… | ||
| The Eyes Are The (Context) Window to The Soul: How Windsurf Gets to Know You | Sam Fertig | 14m |
Windsurf claims the problem isn’t code generation but generating code that fits a large repo and a developer’s personal style, so the system prioritizes… | ||
| Mastering Engineering Flow with Windsurf | Eashan Sinha | 15m |
Windsurf presents an agentic IDE built around the Cascade agent that claims to reduce hallucinations and speed multi-file work by tracking edits, terminal… | ||
| (possible dupe but better sound) What does Enterprise Ready MCP mean? | Tobin South | 14m |
Explains that the hard, unsolved part of scaling MCP is robust authorization and lifecycle management rather than hosting, with specific issues: dynamic client… | ||
| CI in the Era of AI: From Unit Tests to Stochastic Evals | Nathan Sobo | 15m |
Zed is a native Rust editor running at 120 fps that moved from exhaustive deterministic CI to stochastic evals for LLM-driven features; tests include a… | ||
| Fun stories from building OpenRouter and where all this is going | Alex Atallah | 19m |
Company starts in early 2023, evolves from the Window AI Chrome extension into OpenRouter, an aggregator-turned-marketplace that offers a single API for 400… | ||
| Building AI Agents that actually automate Knowledge Work | Jerry Liu | 18m |
A practical blueprint for automating document-heavy workflows: a document MCP server composed of connectors, parser/extractors, and vector/SQL/graph indexes… | ||
| RFT, DPO, SFT: Fine-tuning with OpenAI | Ilan Bigio | 01h 46m |
Covers concrete fine-tuning recipes and outcomes on the OpenAI platform: using the Chat Completions API to SFT function-calling with 120 functions, generating… | ||
| Windsurf everywhere, doing everything, all at once | Kevin Hou | 16m |
Windsurf claims a shared timeline that ingests context from Google Docs, Figma, GitHub, Notion, Linear and runs commands, controls Chrome, opens PRs and… | ||
| Case Study + Deep Dive: Telemedicine Support Agents with LangGraph/MCP | Dan Mason | 01h 56m |
Shows a production telemedicine workflow that replaces many manual console clicks with LLM-driven decisioning, built with LangGraph/MCP, Claude, Node/React… | ||
| Veo 3 for Developers | Paige Bailey | 21m |
Veo 3 claims end-to-end synchronized audio generation paired with video so a single text or image prompt yields dialogue, sound effects, and music without… | ||
| Building Agents with Amazon Nova Act and MCP | Du An Lightfoot | 01h 26m |
Workshop shows building agents that browse and act on the web using Amazon Nova Act, MCP, Bedrock models and the open-source Strands framework. Hands-on demos… | ||
| The Web Browser Is All You Need | Paul Klein IV | 18m |
Claim: the web browser is the default browser MCP server for automating the legacy internet. Evidence: Browserbase runs thousands of headless browsers in the… | ||
| The State of MCP observability: Observable.tools | Alex Volkov, Benjamin Eckel | 17m |
Argues that MCP observability is solvable today by marrying MCP with OpenTelemetry/OTLP, demonstrating a concrete method: extract and inject trace context into… | ||
| Building Protected MCP Servers | Den Delimarsky, Julia Kasper | 20m |
Claims that the new MCP draft spec removes the need for servers to mint tokens by delegating minting to standard authorization servers (Entra ID, Okta… | ||
| The Geopolitics of AI Infrastructure | Dylan Patel | 18m |
Huawei builds a Cloud Matrix 384 that achieves what Nvidia’s DGX H100 Ranger couldn’t: massive NVLink optics scaleup across 384 chips, enabled by TSMC wafers… | ||
| Remote MCPs: What we learned from shipping | John Welsh | 15m |
Anthropic ships remote MCP support for claude.ai and the Anthropic API and standardizes on a single internal protocol by running an MCP gateway that returns an… | ||
| MCP: Origins and Requests For Startups | Theodora Chu | 18m |
MCP launches as an open source protocol last November and quickly goes viral as engineers build integrations and automation then gains adoption in dev tools… | ||
| nothing matches | ||